The case in favour is strong for most workers. A union negotiates collectively, which corrects a genuine imbalance: an individual worker asking for a rise can be refused and replaced, while an employer cannot refuse an entire workforce so easily. Union members in most economies earn measurably more than comparable non-members, and the advantage is largest for lower-paid workers, so unions reduce inequality as well as raising pay. Membership also brings things pay cannot buy — safer working conditions, representation in a disciplinary hearing, negotiated redundancy terms — and those matter most precisely when a worker is most vulnerable.
But "always" is too strong, and several cases genuinely go the other way.
The clearest is the worker who loses their job. A wage pushed above the market level reduces the number of workers employers want to hire, so some members end up unemployed. For them the union has been decisively harmful, and they are invisible in any comparison of union and non-union wages because they are no longer in the industry at all.
There is also the free-rider case. Where negotiated rises apply to everyone in the job regardless of membership, a non-member receives the benefit without paying the subscription or risking lost pay in a strike. Purely financially, that worker does better than a member.
High-performing individuals may also lose. Unions typically negotiate standard rates for a job, which compresses differences between workers, so someone who could have negotiated well above the average on their own merit may be held to the collective rate.
And there are longer-run risks. A union that raises costs substantially may accelerate automation or make the firm uncompetitive, so members' jobs become less secure over time — the gain is real and the job it attaches to is less certain.
My judgement is that most workers most of the time are better off in a union, and that the exceptions follow a pattern rather than being random. The benefit is largest for workers who are hard to replace, low-paid, or in workplaces where conditions matter — there the union wins substantial gains at little cost in jobs. It is smallest, and can be negative, for workers who are easily replaced, because there a wage rise costs many jobs, and for high performers who could do better individually.
That pattern is the same one that determines a union's bargaining power in the first place: it rests on how costly the members are to replace. Where that cost is high, the union delivers; where it is low, it may win little and risk much — which is why union membership is strongest in exactly the occupations where it works best.